On 12/04/2015 03:38 PM, Tolga Cakir wrote: > Am 04.12.2015 um 12:22 schrieb Clément Vuchener: >> On 12/04/2015 07:40 AM, Tolga Cakir wrote: >>> I think we need a decision about what needs to be done in kernel and what is better handled in user-space. I've walked through the Corsair Vengeance K90 driver (hid-corsair.c) and I must say, I'm not really happy about the desicions made in the design. In my opinion, we should only do essential stuff in kernel space and do the rest in user-space. >>> >>> To be more precise about hid-corsair.c (and possibly future gaming keyboard implementations): >>> >>> 1. We should not handle profile switching and exposing a sysfs ABI for profile-number in kernel-space. This has no advantage over keeping track of profiles in user-space. We need to use user-space programs anyway in order to handle macros and profile-sensitive key-handling. Using a keyboard specific ABI for parsing the profile just adds complexity to kernel- and user-space. >>> >>> 2. We should not map false keycodes to keys. hid-corsair.c is using BTN_TRIGGER_HAPPY[N] - this is a huge design flaw in my opinion. Eventhough this might get the job done, I'm questioning the design decision. First of all, we're generally talking about KEYs on a keyboard, not BTNs. Second, they are not TRIGGERs. Third, they are not HAPPY. Also, BTN_TRIGGER_HAPPY[N] is only specified for up to 40 extra keys. What do we do with gaming keyboards, which offer more? The Microsoft Sidewinder X6 has 33 non-standard extra keys for example (agreed, it's less than 40, but still near the maximum). Do we really need to map random linux/input.h keycodes to non-standard keys? >> I choose the TRIGGER_HAPPY buttons because the only use I found for it was extra buttons on some devices. It seems the most appropriate. It would be nice to have special key codes for this kind of keys, I asked about that with my original patch but I did not get any answer so the BTN_TRIGGER_HAPPY stayed there. >> >> They are keys the user may be interested in, I don't see why they should not have key events. > > Hi Clement, > > I agree, but instead of using key events designed for joysticks and gamepads, we should define and use keycodes for our specific use-case. > > How come 40 individual keycodes have been added for a single joystick (Saitek X52 Pro Flight System), whereas macro keys and profile switching keys can be found on hundreds of different gaming keyboards and we have no fitting keycode for that? > > I'm quoting linux/input.h: > > We avoid low common keys in module aliases so they don't get huge. > > Luckily, macro keys and profile switch keys are very common these days on gaming keyboards. So that shouldn't be a problem. > > I think it's time to define KEY_G1 - KEY_G30 and KEY_M1 - KEY_M5, so we can finally have appropiate keycodes for these kind of things. They can be found on nearly every gaming keyboard, so that should be enough to justify them. There is still enough room for additional keycodes. What are you thinking about this? I don't know what would be the correct number of key codes. For Corsair keyboards I know, there is either 18 or 6 G-keys and 3 profile keys (M-keys). So that would be enough for me. My current driver also exposes the MR (macro/mem record) key. But with two different key codes for start and stop. I am not sure it is really useful, I may just remove it when I update the key codes. You should make this proposition to a maintainer. I will update the corsair driver when it is accepted. > >>> I think mapping linux/input.h keycodes to non-standard keys has only one reason: to keep things simple with X11. As Clement mentioned, keycodes above 256 don't work with X11. However, this should not be a reason to hack workarounds into kernel. We can use hidraw / hiddev in user-space instead to catch those events and map them accordingly to our needs (they are designed to be fully programmable keys after all). This adds complexity in user-space, but also adds flexibility and we don't need to break / workaround any HID stuff. >> It does not work with X11 but it works with existing remapping software based on evdev. I don't understand what you mean with "break / workaround", I am remapping *invalid* HID usage code from the keyboard. If I don't do that pressing those keys triggers unwanted key events. > > By "break / workaround" I mean mapping keycodes, which have not been designed for this use-case. BTN suggests, that they are not designed for keyboards in the first place. > >>> 3. We should either have common ABI for gaming keyboards or we shouldn't have them at all (or just in very special cases). This just adds complexity in kernel- and user-space, as lots of keyboards need to be handled their own way. Gaming keyboards have things in common, like LEDs. Creating an ABI for setting extra LEDs is totally legit, but we should define standards, so user-space applications can rely on something. Most keyboards have multiple profile LEDs, 1 recording LED, sometimes 1 gaming mode LED (disabling Windows key) and sometimes keyboard backlight is also adjustable. >>> >>> We could define a common ABI for them: >>> - read-only profile_led_count (or profile_led_max): parse number of profile_leds >>> - rw profile_led_[N]: setting and parsing profile_led_[N] status, 0 and 1 >> Isn't this the same as the current_profile attribute you were arguing against? > > Nope. Exposing an interface to interact with the profile LEDs isn't the same as creating an ABI for the active profile. Turning on profile_led_1 for example does not has to mean, that profile 1 must be active. It just means, that the LED for profile 1 is turned on. > > We should not tie profile to profile_led in kernel space and give user-space applications all flexibility, how to use these LEDs and keeping track of profiles (really, instead of polling / parsing active_profile ABIs, they can easily keep track of them on their own and set LEDs accordingly). I won't be able to do that for the corsair driver, current profile and profile LEDs are tied in the hardware. > >>> - rw recording_led: setting and parsing recording_led status, 0 and 1 >>> - rw gaming_led: setting and parsing gaming_led status, 0 and 1 >>> - read-only backlight_max: parse maximum brightness >>> - rw backlight: setting and parsing backlight intensity, 0 - backlight_max >> LED class devices already do that. > > Perfect! The LED class documentation, defines that LED names should be "devicename:colour:function". I think we should use common function names. For the K90 (and similar keyboards, I expect K40 and K95), I cannot add the profile LEDs as explained above and the "gaming" LED exists but cannot be controlled or read (it is completely managed on the hardware). > > Cheers, > Tolga -- To unsubscribe from this list: send the line "unsubscribe linux-input" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
